Accelerating PostgreSQL Backups with Intel® QuickAssist Technology
Presented by:
Matt Welch
No video of the event yet, sorry!
Problem
PostgreSQL backup operations frequently become critical performance bottlenecks as database compression during pg_dump consumes substantial CPU resources, extending backup windows and impacting concurrent workloads. For organizations managing large PostgreSQL deployments, these bottlenecks directly affect recovery time objectives, operational costs, and system availability.
Solution
This presentation introduces zlib-accel, a transparent software shim layer that offloads PostgreSQL backup compression from CPU cores to Intel® QuickAssist Technology (Intel® QAT) hardware accelerators integrated into 4th Gen Intel® Xeon® processors and later. Leveraging the standard zlib/gzip compression format, this solution functions as a drop-in replacement requiring zero application code changes—simply preload the library and existing pg_dump workflows immediately benefit from hardware acceleration.
Results
Production benchmarking with PostgreSQL 17.4 demonstrates exceptional performance improvements. Intel QAT delivers up to 29.3x throughput increase over standard zlib compression, 7.6x improvement over optimized zlib-ng, and 4.9x acceleration versus modern zstd compression. These dramatic gains are achieved while maintaining competitive compression ratios—Intel QAT produces compressed backups only 2.8% larger than zlib, ensuring storage efficiency remains intact. For Intel® In-Memory Analytics Accelerator (Intel® IAA), throughput increases reach 23.4x over standard zlib with a 5.4% compression ratio tradeoff.
Benefits
Beyond throughput improvements, hardware acceleration substantially reduces CPU cycle consumption during backup operations, freeing compute resources for business-critical database queries and concurrent workloads. The transparent integration approach eliminates development overhead and compatibility concerns, making acceleration immediately accessible to existing PostgreSQL deployments across diverse infrastructure environments.
Takeaways
This session will demonstrate real-world deployment on AWS EC2 i7i instances, present comprehensive performance benchmarking methodology comparing hardware and software compression approaches, discuss practical implementation considerations including driver configuration and system requirements, and provide actionable guidance for database administrators seeking to optimize PostgreSQL backup infrastructure using Intel's hardware acceleration ecosystem.
Speaker Bio
About the Speaker: Matt Welch is a Cloud Software Development Engineer at Intel specializing in database optimization and hardware acceleration technologies. As a co-author of Intel's zlib acceleration white paper, Matt has experience optimizing compression workloads in Apache Kafka and PostgreSQL using Intel QuickAssist Technology and Intel In-Memory Analytics Accelerator.
- Date:
- 2026 April 23 14:00 PDT
- Duration:
- 20 min
- Room:
- Santa Clara (Level C)
- Conference:
- Postgres Conference: 2026
- Language:
- Track:
- Ops
- Difficulty:
- Easy